Nexus.Testing 0.3.0

Nexus

Production-Grade Multi-Agent Runtime For .NET

Build agents, workflows, tools, memory, approvals, protocol bridges, and runtime guardrails in one coherent platform.

Nexus is for teams that want multi-agent systems to behave like engineered software: testable, observable, composable, benchmarked, and ready to operate under real constraints.

Why Teams Pick Nexus

• 🚀 Orchestrate single agents, DAG workflows, parallel branches, fan-out/fan-in graphs, and batched sub-agents
• 🛡️ Enforce approvals, budgets, guardrails, retries, and checkpointing in the runtime instead of hiding control flow inside prompts
• 🧠 Combine tools, memory, messaging, and workflow DSLs without inventing custom orchestration glue
• 🔌 Expose the system through MCP, A2A, AG-UI, and ASP.NET Core hosting endpoints
• 📊 Track token usage, estimated cost, benchmark hot paths, and validate behavior with deterministic tests
• 🧪 Ship with test doubles, mocks, workflow tests, orchestration tests, CLI tests, and live integration coverage
• 🧱 Keep architecture modular across core runtime, protocols, hosting, testing, and standard tools
• ⚙️ Support human-in-the-loop execution, delegated specialist agents, and explicit workflow branching in the same stack
• 📚 Offer recipes, guides, benchmarks, and architecture docs for both humans and LLM-assisted development

Architecture

flowchart TD
    DEFAULTS[Nexus.Defaults]
    HOST[Nexus.Hosting.AspNetCore]
    PROTOCOLS[Nexus.Protocols.Mcp • Nexus.Protocols.A2A • Nexus.Protocols.AgUi]
    DSL[Nexus.Workflows.Dsl]
    ORCH[Nexus.Orchestration • Nexus.Orchestration.Checkpointing]
    LOOP[Nexus.AgentLoop • Nexus.Sessions • Nexus.Compaction]
    RUNTIME[Nexus.Memory • Nexus.Messaging • Nexus.Guardrails • Nexus.Permissions • Nexus.CostTracking • Nexus.Telemetry • Nexus.Auth.OAuth2]
    TOOLS[Nexus.Tools.Standard • Nexus.Commands • Nexus.Skills]
    CONFIG[Nexus.Configuration]
    CORE[Nexus.Core]

    DEFAULTS --> HOST
    DEFAULTS --> LOOP
    HOST --> PROTOCOLS
    PROTOCOLS --> ORCH
    DSL --> ORCH
    ORCH --> LOOP
    LOOP --> RUNTIME
    LOOP --> TOOLS
    RUNTIME --> CORE
    TOOLS --> CORE
    CONFIG --> CORE

Start Here

• Quick Start Guide
• Documentation Index
• Package Index
• LLM Docs
• Installation
• CLI Getting Started
• Quick Start Entry
• Recipe Index
• Examples Index
• Benchmarks README

Recipes

• Existing Provider UI
• Single Agent With Tools
• Chat Session With Memory
• Human-Approved Workflow
• Parallel Sub-Agents And Workflow Fan-Out
• Checkpointed Recovery Workflow
• Tool-Only Worker Agent
• Cost-Aware Batch Processing
• Task System + Graph Brain

Runnable Scenario Examples

• Single Agent With Tools Example
• Chat Session With Memory Example
• Human-Approved Workflow Example
• Parallel Sub-Agents And Workflow Fan-Out Example
• Chat Editing With Diff And Revert Example

Guides

• Quick Start
• Orchestration
• Sub-Agents
• Performance And Benchmarking
• Production Hardening
• CI And Quality Gates
• Workflow Patterns And Anti-Patterns
• External Brain & Task System
• Memory
• Guardrails
• Permissions
• Cost Tracking
• Messaging
• Checkpointing
• Workflows DSL
• Protocols
• Telemetry
• Auth
• Testing
• Middleware

Architecture Docs

• Architecture Overview
• Core Engine

API Docs

• Package Index
• Nexus.Core
• Nexus.AgentLoop
• Nexus.Auth.OAuth2
• Nexus.Commands
• Nexus.Compaction
• Nexus.Configuration
• Nexus.Defaults
• Nexus.Orchestration
• Nexus.Orchestration.Checkpointing
• Nexus.Memory
• Nexus.Guardrails
• Nexus.Permissions
• Nexus.CostTracking
• Nexus.Messaging
• Nexus.Sessions
• Nexus.Skills
• Nexus.Telemetry
• Nexus.Tools.Standard
• Nexus.Workflows.Dsl
• Nexus.Protocols
• Nexus.Protocols.Mcp
• Nexus.Protocols.A2A
• Nexus.Protocols.AgUi
• Nexus.Hosting.AspNetCore
• Nexus.Testing

LLM Docs

• LLM Docs Index
• Runtime Map
• Agent Loop
• Workflows DSL
• Tools And Sub-Agents
• Testing And Benchmarks
• Glossary

Examples Docs

• Minimal Agent
• Multi-Agent Graph
• Chat Editing With Diff And Revert
• Nexus CLI
• Examples Index

Project Structure

• src: Core runtime, orchestration, checkpointing, memory, messaging, guardrails, permissions, cost tracking, telemetry, auth, protocols, standard tools, workflow DSL, hosting, and testing utilities
• examples: CLI, minimal setup, multi-agent examples, and runnable recipe apps
• tests: Unit, integration, live, CLI, orchestration, workflow, and runtime tests
• benchmarks: Workflow and sub-agent benchmark suite
• docs: Guides, recipes, LLM docs, API references, architecture, and getting started material

Test Coverage And Test Count

• 389 tests passed in the latest full solution run across 25 test projects
• Coverage metrics are collected via CI and updated with each release

Benchmark Results

• benchmarks/Nexus.Benchmarks contains the workflow and sub-agent runtime benchmark suite
• Latest measured results on the current workstation:
• CompileWorkflow: 1.788 μs mean, 5.15 KB allocated
• ExecuteWorkflow: 92.894 μs mean, 39.17 KB allocated
• RunParallelSubAgents: 3.670 μs mean, 8.2 KB allocated
• Full benchmark reports are written to BenchmarkDotNet.Artifacts/results

Credits

• Built on .NET 10 and Microsoft.Extensions.DependencyInjection
• AI abstraction layer powered through Microsoft.Extensions.AI
• Workflow serialization support via YamlDotNet
• Benchmarking via BenchmarkDotNet
• Testing with xUnit and FluentAssertions
• CLI presentation powered by Spectre.Console

License

MIT